/*
* thread to combine salvager child logs
* back into the main salvageserver log
*/
static void *
SalvageLogCleanupThread(void * arg)
{
struct log_cleanup_node * cleanup;
assert(pthread_mutex_lock(&worker_lock) == 0);
while (1) {
while (queue_IsEmpty(&log_cleanup_queue)) {
assert(pthread_cond_wait(&log_cleanup_queue.queue_change_cv, &worker_lock) == 0);
}
while (queue_IsNotEmpty(&log_cleanup_queue)) {
cleanup = queue_First(&log_cleanup_queue, log_cleanup_node);
queue_Remove(cleanup);
assert(pthread_mutex_unlock(&worker_lock) == 0);
SalvageLogCleanup(cleanup->pid);
free(cleanup);
assert(pthread_mutex_lock(&worker_lock) == 0);
}
}
assert(pthread_mutex_unlock(&worker_lock) == 0);
return NULL;
}
/*
* thread to combine salvager child logs
* back into the main salvageserver log
*/
static void *
SalvageLogCleanupThread(void * arg)
{
struct log_cleanup_node * cleanup;
MUTEX_ENTER(&worker_lock);
while (1) {
while (queue_IsEmpty(&log_cleanup_queue)) {
CV_WAIT(&log_cleanup_queue.queue_change_cv, &worker_lock);
}
while (queue_IsNotEmpty(&log_cleanup_queue)) {
cleanup = queue_First(&log_cleanup_queue, log_cleanup_node);
queue_Remove(cleanup);
MUTEX_EXIT(&worker_lock);
SalvageLogCleanup(cleanup->pid);
free(cleanup);
MUTEX_ENTER(&worker_lock);
}
}
MUTEX_EXIT(&worker_lock);
return NULL;
}
/**
* dequeue a node from a list object.
*
* @param[in] list list object
* @param[out] node_out address in which to store node object pointer
* @param[in] state new node state
* @param[in] block permit blocking on cv if asserted
*
* @return operation status
* @retval 0 success
* @retval EWOULDBLOCK block not asserted and nothing to dequeue
* @retval EINTR blocking wait interrupted by list shutdown
*
* @post node object returned with node lock held and new state set
*
* @internal
*/
static int
_afs_wq_node_list_dequeue(struct afs_work_queue_node_list * list,
struct afs_work_queue_node ** node_out,
afs_wq_work_state_t state,
int block)
{
int ret = 0;
struct afs_work_queue_node * node;
MUTEX_ENTER(&list->lock);
if (list->shutdown) {
*node_out = NULL;
ret = EINTR;
goto done_sync;
}
if (!block && queue_IsEmpty(&list->list)) {
*node_out = NULL;
ret = EWOULDBLOCK;
goto done_sync;
}
while (queue_IsEmpty(&list->list)) {
if (list->shutdown) {
*node_out = NULL;
ret = EINTR;
goto done_sync;
}
CV_WAIT(&list->cv, &list->lock);
}
*node_out = node = queue_First(&list->list, afs_work_queue_node);
MUTEX_ENTER(&node->lock);
queue_Remove(node);
node->qidx = AFS_WQ_NODE_LIST_NONE;
_afs_wq_node_state_change(node, state);
done_sync:
MUTEX_EXIT(&list->lock);
return ret;
}
/**
* append to a node list object.
*
* @param[in] list list object
* @param[in] node node object
* @param[in] state new node state
*
* @return operation status
* @retval 0 success
* @retval AFS_WQ_ERROR raced to enqueue node
*
* @pre
* - node lock held
* - node is not on a list
* - node is either not busy, or it is marked as busy by the calling thread
*
* @post
* - enqueued on list
* - node lock dropped
*
* @internal
*/
static int
_afs_wq_node_list_enqueue(struct afs_work_queue_node_list * list,
struct afs_work_queue_node * node,
afs_wq_work_state_t state)
{
int code, ret = 0;
if (node->qidx != AFS_WQ_NODE_LIST_NONE) {
/* raced */
ret = AFS_WQ_ERROR;
goto error;
}
/* deal with lock inversion */
code = MUTEX_TRYENTER(&list->lock);
if (!code) {
/* contended */
_afs_wq_node_state_change(node, AFS_WQ_NODE_STATE_BUSY);
MUTEX_EXIT(&node->lock);
MUTEX_ENTER(&list->lock);
MUTEX_ENTER(&node->lock);
/* assert state of the world (we set busy, so this should never happen) */
osi_Assert(queue_IsNotOnQueue(node));
}
if (list->shutdown) {
ret = AFS_WQ_ERROR;
goto error_unlock;
}
osi_Assert(node->qidx == AFS_WQ_NODE_LIST_NONE);
if (queue_IsEmpty(&list->list)) {
/* wakeup a dequeue thread */
CV_SIGNAL(&list->cv);
}
queue_Append(&list->list, node);
node->qidx = list->qidx;
_afs_wq_node_state_change(node, state);
error_unlock:
MUTEX_EXIT(&node->lock);
MUTEX_EXIT(&list->lock);
error:
return ret;
}
/* Optimization for unmarshalling 32 bit integers */
int
rx_ReadProc32(struct rx_call *call, afs_int32 * value)
{
int bytes;
int tcurlen;
int tnLeft;
char *tcurpos;
SPLVAR;
/* Free any packets from the last call to ReadvProc/WritevProc */
if (!queue_IsEmpty(&call->iovq)) {
#ifdef RXDEBUG_PACKET
call->iovqc -=
#endif /* RXDEBUG_PACKET */
rxi_FreePackets(0, &call->iovq);
}
/*
* Most common case, all of the data is in the current iovec.
* We are relying on nLeft being zero unless the call is in receive mode.
*/
tcurlen = call->curlen;
tnLeft = call->nLeft;
if (!call->error && tcurlen >= sizeof(afs_int32)
&& tnLeft >= sizeof(afs_int32)) {
tcurpos = call->curpos;
memcpy((char *)value, tcurpos, sizeof(afs_int32));
call->curpos = tcurpos + sizeof(afs_int32);
call->curlen = (u_short)(tcurlen - sizeof(afs_int32));
call->nLeft = (u_short)(tnLeft - sizeof(afs_int32));
if (!call->nLeft && call->currentPacket != NULL) {
/* out of packet. Get another one. */
rxi_FreePacket(call->currentPacket);
call->currentPacket = (struct rx_packet *)0;
}
return sizeof(afs_int32);
}
NETPRI;
bytes = rxi_ReadProc(call, (char *)value, sizeof(afs_int32));
USERPRI;
return bytes;
}
int
rx_ReadProc(struct rx_call *call, char *buf, int nbytes)
{
int bytes;
int tcurlen;
int tnLeft;
char *tcurpos;
SPLVAR;
/* Free any packets from the last call to ReadvProc/WritevProc */
if (!queue_IsEmpty(&call->iovq)) {
#ifdef RXDEBUG_PACKET
call->iovqc -=
#endif /* RXDEBUG_PACKET */
rxi_FreePackets(0, &call->iovq);
}
/*
* Most common case, all of the data is in the current iovec.
* We are relying on nLeft being zero unless the call is in receive mode.
*/
tcurlen = call->curlen;
tnLeft = call->nLeft;
if (!call->error && tcurlen > nbytes && tnLeft > nbytes) {
tcurpos = call->curpos;
memcpy(buf, tcurpos, nbytes);
call->curpos = tcurpos + nbytes;
call->curlen = tcurlen - nbytes;
call->nLeft = tnLeft - nbytes;
if (!call->nLeft && call->currentPacket != NULL) {
/* out of packet. Get another one. */
rxi_FreePacket(call->currentPacket);
call->currentPacket = (struct rx_packet *)0;
}
return nbytes;
}
NETPRI;
bytes = rxi_ReadProc(call, buf, nbytes);
USERPRI;
return bytes;
}
/* Add the indicated event (function, arg) at the specified clock time */
struct rxevent *
rxevent_Post(struct clock * when, void (*func)(), void *arg, void *arg1)
/* when - When event should happen, in clock (clock.h) units */
{
struct rxevent *ev, *qe, *qpr;
#ifdef RXDEBUG
if (Log) {
struct clock now;
clock_GetTime(&now);
fprintf(Log, "%ld.%ld: rxevent_Post(%ld.%ld, %p, %p)\n",
now.sec, now.usec, when->sec, when->usec, func, arg);
}
#endif
#if defined(AFS_SGIMP_ENV)
ASSERT(osi_rxislocked());
#endif
/*
* If we're short on free event entries, create a block of new ones and
* add them to the free queue
*/
if (queue_IsEmpty(&rxevent_free)) {
int i;
#if defined(AFS_AIX32_ENV) && defined(KERNEL)
ev = (struct rxevent *) rxi_Alloc(sizeof(struct rxevent));
queue_Append(&rxevent_free, &ev[0]), rxevent_nFree++;
#else
ev = (struct rxevent *) osi_Alloc(sizeof(struct rxevent) *
rxevent_allocUnit);
xsp = xfreemallocs;
xfreemallocs = (struct xfreelist *) ev;
xfreemallocs->next = xsp;
for (i = 0; i < rxevent_allocUnit; i++)
queue_Append(&rxevent_free, &ev[i]), rxevent_nFree++;
#endif
}
/* Grab and initialize a new rxevent structure */
ev = queue_First(&rxevent_free, rxevent);
queue_Remove(ev);
rxevent_nFree--;
/* Record user defined event state */
ev->eventTime = *when;
ev->func = func;
ev->arg = arg;
ev->arg1 = arg1;
rxevent_nPosted += 1; /* Rather than ++, to shut high-C up
* regarding never-set variables */
/*
* Locate a slot for the new entry. The queue is ordered by time, and we
* assume that a new entry is likely to be greater than a majority of the
* entries already on the queue (unless there's very few entries on the
* queue), so we scan it backwards
*/
for (queue_ScanBackwards(&rxevent_queue, qe, qpr, rxevent)) {
if (clock_Ge(when, &qe->eventTime)) {
queue_InsertAfter(qe, ev);
return ev;
}
}
/* The event is to expire earlier than any existing events */
queue_Prepend(&rxevent_queue, ev);
if (rxevent_ScheduledEarlierEvent)
(*rxevent_ScheduledEarlierEvent) (); /* Notify our external
* scheduler */
return ev;
}
/* rxi_WritevProc -- internal version.
*
* Send buffers allocated in rxi_WritevAlloc.
*
* LOCKS USED -- called at netpri.
*/
int
rxi_WritevProc(struct rx_call *call, struct iovec *iov, int nio, int nbytes)
{
struct rx_packet *cp = NULL;
#ifdef RX_TRACK_PACKETS
struct rx_packet *p, *np;
#endif
int nextio;
int requestCount;
struct rx_queue tmpq;
#ifdef RXDEBUG_PACKET
u_short tmpqc;
#endif
requestCount = nbytes;
nextio = 0;
MUTEX_ENTER(&call->lock);
if (call->error) {
call->mode = RX_MODE_ERROR;
} else if (call->mode != RX_MODE_SENDING) {
call->error = RX_PROTOCOL_ERROR;
}
#ifdef AFS_GLOBAL_RXLOCK_KERNEL
rxi_WaitforTQBusy(call);
#endif /* AFS_GLOBAL_RXLOCK_KERNEL */
cp = call->currentPacket;
if (call->error) {
call->mode = RX_MODE_ERROR;
MUTEX_EXIT(&call->lock);
if (cp) {
#ifdef RX_TRACK_PACKETS
cp->flags &= ~RX_PKTFLAG_CP;
cp->flags |= RX_PKTFLAG_IOVQ;
#endif
queue_Prepend(&call->iovq, cp);
#ifdef RXDEBUG_PACKET
call->iovqc++;
#endif /* RXDEBUG_PACKET */
call->currentPacket = (struct rx_packet *)0;
}
#ifdef RXDEBUG_PACKET
call->iovqc -=
#endif /* RXDEBUG_PACKET */
rxi_FreePackets(0, &call->iovq);
return 0;
}
/* Loop through the I/O vector adjusting packet pointers.
* Place full packets back onto the iovq once they are ready
* to send. Set RX_PROTOCOL_ERROR if any problems are found in
* the iovec. We put the loop condition at the end to ensure that
* a zero length write will push a short packet. */
nextio = 0;
queue_Init(&tmpq);
#ifdef RXDEBUG_PACKET
tmpqc = 0;
#endif /* RXDEBUG_PACKET */
do {
if (call->nFree == 0 && cp) {
clock_NewTime(); /* Bogus: need new time package */
/* The 0, below, specifies that it is not the last packet:
* there will be others. PrepareSendPacket may
* alter the packet length by up to
* conn->securityMaxTrailerSize */
hadd32(call->bytesSent, cp->length);
rxi_PrepareSendPacket(call, cp, 0);
queue_Append(&tmpq, cp);
#ifdef RXDEBUG_PACKET
tmpqc++;
#endif /* RXDEBUG_PACKET */
cp = call->currentPacket = (struct rx_packet *)0;
/* The head of the iovq is now the current packet */
if (nbytes) {
if (queue_IsEmpty(&call->iovq)) {
MUTEX_EXIT(&call->lock);
call->error = RX_PROTOCOL_ERROR;
#ifdef RXDEBUG_PACKET
tmpqc -=
#endif /* RXDEBUG_PACKET */
rxi_FreePackets(0, &tmpq);
return 0;
}
cp = queue_First(&call->iovq, rx_packet);
queue_Remove(cp);
#ifdef RX_TRACK_PACKETS
cp->flags &= ~RX_PKTFLAG_IOVQ;
#endif
#ifdef RXDEBUG_PACKET
call->iovqc--;
#endif /* RXDEBUG_PACKET */
#ifdef RX_TRACK_PACKETS
cp->flags |= RX_PKTFLAG_CP;
#endif
call->currentPacket = cp;
call->nFree = cp->length;
call->curvec = 1;
call->curpos =
(char *)cp->wirevec[1].iov_base +
call->conn->securityHeaderSize;
call->curlen =
cp->wirevec[1].iov_len - call->conn->securityHeaderSize;
}
}
if (nbytes) {
/* The next iovec should point to the current position */
if (iov[nextio].iov_base != call->curpos
|| iov[nextio].iov_len > (int)call->curlen) {
call->error = RX_PROTOCOL_ERROR;
MUTEX_EXIT(&call->lock);
if (cp) {
#ifdef RX_TRACK_PACKETS
cp->flags &= ~RX_PKTFLAG_CP;
#endif
queue_Prepend(&tmpq, cp);
#ifdef RXDEBUG_PACKET
tmpqc++;
#endif /* RXDEBUG_PACKET */
cp = call->currentPacket = (struct rx_packet *)0;
}
#ifdef RXDEBUG_PACKET
tmpqc -=
#endif /* RXDEBUG_PACKET */
rxi_FreePackets(0, &tmpq);
return 0;
}
nbytes -= iov[nextio].iov_len;
call->curpos += iov[nextio].iov_len;
call->curlen -= iov[nextio].iov_len;
call->nFree -= iov[nextio].iov_len;
nextio++;
if (call->curlen == 0) {
if (++call->curvec > cp->niovecs) {
call->nFree = 0;
} else {
call->curpos = (char *)cp->wirevec[call->curvec].iov_base;
call->curlen = cp->wirevec[call->curvec].iov_len;
}
}
}
} while (nbytes && nextio < nio);
/* Move the packets from the temporary queue onto the transmit queue.
* We may end up with more than call->twind packets on the queue. */
#ifdef RX_TRACK_PACKETS
for (queue_Scan(&tmpq, p, np, rx_packet))
{
p->flags |= RX_PKTFLAG_TQ;
}
#endif
if (call->error)
call->mode = RX_MODE_ERROR;
queue_SpliceAppend(&call->tq, &tmpq);
if (!(call->flags & (RX_CALL_FAST_RECOVER | RX_CALL_FAST_RECOVER_WAIT))) {
rxi_Start(0, call, 0, 0);
}
/* Wait for the length of the transmit queue to fall below call->twind */
while (!call->error && call->tnext + 1 > call->tfirst + (2 * call->twind)) {
clock_NewTime();
call->startWait = clock_Sec();
#ifdef RX_ENABLE_LOCKS
CV_WAIT(&call->cv_twind, &call->lock);
#else
call->flags |= RX_CALL_WAIT_WINDOW_ALLOC;
osi_rxSleep(&call->twind);
#endif
call->startWait = 0;
}
/* cp is no longer valid since we may have given up the lock */
cp = call->currentPacket;
if (call->error) {
call->mode = RX_MODE_ERROR;
call->currentPacket = NULL;
MUTEX_EXIT(&call->lock);
if (cp) {
#ifdef RX_TRACK_PACKETS
cp->flags &= ~RX_PKTFLAG_CP;
#endif
rxi_FreePacket(cp);
}
return 0;
}
MUTEX_EXIT(&call->lock);
return requestCount - nbytes;
}
